HPC, HTC and Cloud

HPC, HTC, and Cloud Computing are modern answers to “too much data”.

HPC, HTC, and Cloud Computing are modern answers to “too much data”. As given by the facts listed here:

It is pretty easy to immerse oneself in data and quickly run out of the necessary horsepower to complete a project in a timely manner. Today, we scale our compute.


What were you doing during the debates??

Market by President’s Party


Returns averaged by party and corrected for plans for tax rate:

party NASDAQ NYA SP DJIA
Dem 0.20 0.12 0.14 0.14
Rep 0.09 0.06 0.07 0.06
Party NASDAQ-corr NYA-corr SP-corr DJIA-corr
Dem 0.17 0.10 0.12 0.12
Rep 0.06 0.04 0.04 0.04

HPC

HPC, high performance computing, is generally characterized by low latency, high bandwith connections between many identical nodes organized as a single cluster. HTC, replace the low latency connection with standard ethernet and Cloud abstracts the hardware away with virtual machines and removes the generalization of nodes being identical. Here we will talk through high performance computing as it is configured and practiced at Virginia Tech.

Computing in shared cluster environments requires both new terminology and a new way to think about computing. Just pressing GO on a GUI can land you in some trouble.


What does a cluster look like?

HPC system diagram

There are a couple of items to note.


HPC cluster roll up Above is old news!! TinkerCliffs: 1 AMD node = 2 sockets => 64 cores/socket => 128 cores/node 308 nodes => 40448 (+ 1536 cores – 96 cores/node on 16 Intel nodes)


This leads us to a number of terms:


GUT check

Can everyone log in to https://ood.arc.vt.edu Remember, VPN must be on!

Break while answering the above …